Package org.moeaframework.analysis.sensitivity

Class SobolSensitivityAnalysis

java.lang.Object

org.moeaframework.analysis.sensitivity.SobolSensitivityAnalysis

All Implemented Interfaces:

SensitivityAnalysis<SobolSensitivityAnalysis.SobolSensitivityResult>

public class SobolSensitivityAnalysis
extends Object
implements SensitivityAnalysis<SobolSensitivityAnalysis.SobolSensitivityResult>

Global sensitivity analysis of blackbox model output using Saltelli's improved Sobol' global variance decomposition procedure.

1. When requesting N samples, the Saltelli sampling strategy generates N * (2 * P + 2) actual samples, where P is the number of parameters being analyzed.
2. Negative sensitivity values can occur and typically coincide with large confidence intervals. Increasing the sample size can help, but this generally means the sensitivities are near zero.

This code was derived and translated from the C code used in the Tang et al. (2007) study cited below.

References:

1. Tang, Y., Reed, P., Wagener, T., and van Werkhoven, K., "Comparing Sensitivity Analysis Methods to Advance Lumped Watershed Model Identification and Evaluation," Hydrology and Earth System Sciences, vol. 11, no. 2, pp. 793-817, 2007.
2. Saltelli, A., et al. "Global Sensitivity Analysis: The Primer." John Wiley & Sons Ltd, 2008.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
class	SobolSensitivityAnalysis.SobolSensitivityResult	Computes the results of the Sobol' global variance decomposition procedure.

Constructor Summary

Constructors

Constructor	Description
SobolSensitivityAnalysis(ParameterSet parameterSet, int N)	Constructs a new Sobol' global variance decomposition instance.
SobolSensitivityAnalysis(ParameterSet parameterSet, int N, int resamples)	Constructs a new Sobol' global variance decomposition instance.

Method Summary

Modifier and Type	Method	Description
SobolSensitivityAnalysis.SobolSensitivityResult	evaluate(double[] responses)	Evaluates the model responses associated with each sample, returning the sensitivity results.
Samples	generateSamples()	Generates and returns the samples required by this sensitivity analysis method.
ParameterSet	getParameterSet()	Returns the parameter set associated with this sensitivity analysis.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface org.moeaframework.analysis.sensitivity.SensitivityAnalysis

evaluate

Constructor Details

SobolSensitivityAnalysis

public SobolSensitivityAnalysis(ParameterSet parameterSet,
 int N)

Constructs a new Sobol' global variance decomposition instance.

Parameters:

parameterSet - the parameters being analyzed

N - the number of samples

SobolSensitivityAnalysis

public SobolSensitivityAnalysis(ParameterSet parameterSet,
 int N,
 int resamples)

Constructs a new Sobol' global variance decomposition instance.

Parameters:

parameterSet - the parameters being analyzed

N - the number of samples

resamples - the number of resamples used to bootstrap the 50% confidence intervals

Method Details

getParameterSet

public ParameterSet getParameterSet()

Description copied from interface: SensitivityAnalysis

Returns the parameter set associated with this sensitivity analysis.

Specified by:

getParameterSet in interface SensitivityAnalysis<SobolSensitivityAnalysis.SobolSensitivityResult>

Returns:

the parameter set

generateSamples

public Samples generateSamples()

Description copied from interface: SensitivityAnalysis

Generates and returns the samples required by this sensitivity analysis method. Typically, but not always, these samples are random or pseudo-random.

Specified by:

generateSamples in interface SensitivityAnalysis<SobolSensitivityAnalysis.SobolSensitivityResult>

Returns:

the samples

evaluate

public SobolSensitivityAnalysis.SobolSensitivityResult evaluate(double[] responses)

Description copied from interface: SensitivityAnalysis

Evaluates the model responses associated with each sample, returning the sensitivity results. The order of responses must match the order of the generated samples!

Specified by:

evaluate in interface SensitivityAnalysis<SobolSensitivityAnalysis.SobolSensitivityResult>

Parameters:

responses - the model responses

Returns:

the sensitivity results